Resolving the datacenters’ insatiable appetite for energy

As datacenters’ capacity increases, so does energy consumption. The need is to remain cognizant of the effects, and keep pushing for more sustainable approaches to design, manufacturing, and consumption.

Datacenter capacity in India is expected to double from 870 MW last fiscal, to 1800 MW by fiscal 2025, translating into a required investment of Rs 40,000 crore.

The corporate embrace of advanced technologies and digital infrastructure, increasing use of smart devices by individuals, rollout of 5G services, the sharp increase in cloud adoption, and digital transformation of businesses are leading to a massive spurt in data, creating huge demand for datacenters.

The rise of artificial intelligence (AI) is also anticipated to provide significant growth in datacenters. Tech disruptions, such as the metaverse, have currently gained traction, and datacenters will play a huge role in creating a digital environment that is more convincing, immersive, and high on processing power. The government’s on-shoring of data of Indian citizens itself will create an enormous opportunity.

In the past year, several companies have announced expansion plans of their datacenter capacities in India. Yotta, with two datacenters in India, plans to deliver about 1000 MW of capacity by 2030, at an investment of Rs 60,000 crore. NTT has announced a budget of USD 2 billion toward India investments till 2024. Equinix and ST Telemedia have also announced expansion of their datacenter capacities in India.

Digital Edge (Singapore) Holdings Pte. Ltd., under the brand name Digital Edge DC, has entered into a partnership with the National Investment and Infrastructure Fund and AGP DC InvestCo Pte. Ltd. to develop a pan-India portfolio of hyperscale datacenters. The partnership’s first project is a USD 2-billion investment, being set up between Thane and Navi Mumbai, aimed at developing a 300-MW hyperscale facility, one of the largest datacenters in the country.

AdaniConneX, having opened its first hyperscale datacenter at Chennai in November 2022, is investing capital into this sector over the next decade, with a mission to build over 1-GW green datacenter platform. The company is building hyperscale campuses at Mumbai, Navi Mumbai, Noida, Pune, Kolkata, Bhubaneshwar, Bengaluru, Hyderabad, and Vizag. AdaniConneX is also developing distributed Edge Datacenters and Far Edge facilities in Tier-II and Tier-III markets.

Datacenters need electricity to run their equipment and to keep the machines cool. They are estimated to be responsible for up to 3 percent of global electricity consumption today, projected to touch 4 percent by 2030. Each facility consumes in the order of 20 MW up to 40 MW of power, and most of the time datacenters are running at 100-percent utilization, as all the processors are being kept busy. A 20-MW facility probably draws 20 MW fairly consistently – enough to power roughly 16,000 households – computing as much as it can to amortize the costs of the datacenter, its servers, and power delivery systems. Electricity could account for 45–50 percent of the operating expense of a datacenter.

The water usage effectiveness (WUE) of an average datacenter using evaporative cooling systems is 1.8 L per kWh. That type of datacenter can consume 3–5 million gallons of water per day – similar to the capacity used by a city of 30,000–50,000 people.

The advances in chip design and manufacturing that limited server power consumption through the first decade and a half of the 2000s reached their limits in recent years, and a spike in the amount of energy servers use has followed server power consumption has increased by 266 percent since 2017.

The industry is continuing to take steps to self-monitor and moderate. More companies have pledged to meet global climate targets and cut carbon emissions, accelerating the push toward carbon neutrality.

Datacenter operators recognize that changes need to be made and that new energy business models will be key to how datacenters view and manage their power needs. The good news is that by 2050, The International Renewable Energy Agency (IRENA) estimates that 90 percent of the world’s electricity can and should come from renewable energy. The bad news, however, is the intermittent nature of renewable energy poses flexibility challenges not only to datacenters but also to the energy grid itself. And all it takes is a large spike in consumption or drop in production to severely destabilize the grid. For that reason, a datacenter cannot solely rely on renewable energy. The services in which datacenters can offer the grid vary, from energy generation and export, to storage and frequency containment.

There is sharper focus on an optimum mix of grid power and renewables. Energy from the grid still burns at an alarming 70 percent of fossil fuel, proving to be a challenge for the planet. The share of renewables in datacenter power consumption is expected to increase to ~35–40 percent by fiscal 2025 from less than 15 percent now. Renewable power being cheaper will improve the operating margins of the sector by ~200–300 basis points by fiscal 2025, and help sustain project’s returns on capital employed at 13–15 percent.

The diesel generator has long been an imperfect but inescapable piece of the datacenter ecosystem. It represents stored energy that largely goes unused while still requiring maintenance or fuel replacement after periods of inactivity. Then, when pressed into service, generators produce carbon emissions operators are desperately trying to avoid. Already, some organizations are relying on batteries for longer load support – up to five minutes in some cases – and even designing their datacenters with minimal generator capacity.

These are transitional steps to minimize the role of the generator as the industry searches for other options – including new battery technologies – for extended backup power. Experts anticipate a preferred alternative will emerge – specifically hydrogen fuel cells. These fuel cells will function much like a generator at first, providing momentary load support, and eventually hold promise for sustained or even continuous operation.

Higher densities alter thermal strategies. After years of relatively static rack densities, datacenter operators are increasingly requesting higher-density racks. And, with more power draw comes more waste heat that needs to be removed from the rack and eventually the white space. When racks consumed up to 20 kW, air-based cooling methodologies could be relied on to keep the IT hardware operating safely and efficiently. But as some racks start to exceed 30 kW or more, new cooling approaches need to be used.

Although air-based cooling options exist for racks drawing more than 20 kW, they are often cumbersome to install and maintain effectively, essentially passing the point of diminishing returns in terms of cooling capacity. As such, datacenters are now cautiously looking toward liquid cooling for their new facility projects.

While liquid cooling is not a new technology, the early wave of successful, efficient, problem-free deployments in high-density environments has provided proof of concept that will boost adoption in the coming year. There are essentially liquid cooling options for enterprise-grade IT hardware. There are essentially two main categories of liquid cooling – direct-to-chip liquid cooling (sometimes called conductive or cold plate liquid cooling) and immersive liquid cooling. The addition of direct-to-chip cooling to new OCP and Open19 standards will only accelerate this trend. Direct chip liquid cooling can offer some of the lowest PUE possible, as the temperature at which they operate means that no mechanical or adiabatic cooling would be required.

Far from mainstream, liquid cooling is positioning itself as the cooling solution for high-performance computing. Its mainstream adoption will, however, depend on advances in technology and chip designs. Retrofitting already existing datacenters is costly for some forms of liquid cooling, while the weight of immersion tanks makes it impractical for many current raised-floor facilities.

More compute power may seem like it will result in significantly more power usage, but in fact, as it uses and produces higher temperatures, this leads to greater efficiency – not only in PUE but also in other resources, such as water. Higher-powered compute often uses greater intelligence in their software, so there is an opportunity to innovate to lower the PUE further. In the future, it may even mean that this kind of software could enable the removal of generators or UPSs completely.

A datacenter’s carbon performance is broadly a function of the energy mix in the location in which it is operating. There are some exceptions, where operators take the responsibility to generate power on-site, using renewables or gas, but largely speaking local grids power datacenters.

To reduce the carbon footprint, companies are introducing methods, which act as a yardstick to an environmental sustainability metric for datacenters. For example, guides, such as the White Paper 67, prove to be a game changer for sustainable datacenters. By introducing the required metrics, datacenter service providers and companies can report on energy usage, greenhouse gas emissions, and use of water. Other metrics that include measured values, such as carbon usage effectiveness and land and biodiversity, will help a diverse set of companies move toward sustainability at a faster pace.

With carbon emissions on rise, and countries taking big steps toward a sustainable tomorrow, green datacenters and sustainability metrics are the need of the hour. The global green datacenter market is expected to grow from USD 58.06 billion in 2022 to USD 120.26 billion in 2026 at a CAGR of 19.97 percent.

And it is not just about the hardware; software is key. It not only provides intelligent operational insights and actionable data but also brings policy-based elements of automation and control. Effectively rolling DCIM and EPMS into one digital platform, 3D monitoring of the entire datacenter environment, pulling in data from all essential assets within the facility needs to be provided. This gives operators clear visibility and insights into their critical functions and operational data sets, enabling them to increase levels of control and automation.

The major factors guiding the efficiency and sustainability efforts of datacenters can be identified as need to increase renewables, improving energy storage, providing power intelligence insights to more people – from technicians to the c-suite and promoting that story to regulators, markets, and customers.

The datacenter industry is a major buyer of power purchase agreements (PPA) for renewable energy, and this has a significant impact on the energy mix. In 2021, Amazon and Microsoft were the two largest corporate buyers of renewable energy through PPA; to a degree, the datacenter industry is helping to drive de-carbonization by underwriting a significant proportion of grid-scale, carbon-free energy for industry. This can catalyze a whole ecosystem at a national level, and also demonstrates to the broader industrial base that critical loads can reliably move to renewables.

Taking the case study of Microsoft further, the company has been demonstrating how datacenters can conserve power, reduce emissions, and even contribute energy back to the grid.

In Finland, waste heat from two new datacenters will contribute to the district heating system that provides warmth to more than 250,000 people in winter. The Microsoft datacenter region in Sweden uses rainwater and outside air to cool servers, while using the heat they produce to keep work areas warm for employees. Also in Sweden, Microsoft is piloting batteries to displace diesel generators as backup systems.

That means fossil-fuel burning power plants will be needed less often to maintain steady power and cutting emissions and fuel costs. The great thing about the project in Ireland was that those batteries were already there. What it required was providing that digital layer of intelligence to determine what the grid needs to help balance the frequency on the system.

Those assets, which are ubiquitous in datacenters, are all over the world. And it creates a huge opportunity to be able to see the datacenter as something more than a consumer of energy, but also a producer and a partner to grid operators to improve reliability and ultimately the energy transition.

There is huge potential in the development of digital tools to help grid operators shift loads during periods of high demand. IoT and AI could help create energy efficiency in a variety of ways. AI can be used for everything from smoothing out supply-chain issues to creating more accurate local weather forecasting to helping providers find ways to capture more energy. While AI and machine learning will add to demand for cloud computing, those advanced tools are likely to be essential to accelerate the energy transition.

Gains in power efficiency have stalled during the past decade

Growing adoption of renewable energy – and higher levels of infrastructure redundancy at the IT level – are also leading to new design best practices in datacenters. Battery energy storage systems (BESS) are replacing diesel gensets as short-term backup power supply, for example. With energy markets increasingly interconnected, datacenters adopting BESS can generate unprecedented revenues by running in island mode in case of outages (without emitting carbon dioxide from diesel generators) or stabilizing the frequency of the grid.

Building elasticity into datacenter design is also a key factor in reducing the sector’s energy consumption and cost. Through intelligent design, instrumentation, control and automation, a datacenter can enable and disable capacity when it is needed, rather than constantly running circuits and networks with no work to do. In addition to preventing the over-provision of infrastructure, this is crucial at times of the year, when additional capacity is needed to cover short peaks in demand. Black Friday and the Christmas period are a perfect example; with better mechanical, electrical, and automation design, we can dynamically reduce or increase datacenter infrastructure resources, enabling them to reliably run closer to their capacity for short periods of time.

The future. Digitization and the rise of internet access have delivered significant benefits to humanity, but the sustainability impact must be addressed. The mission now is to continually and steadily drive that impact toward zero. The datacenter industry has led by example by willingly investing in de-carbonization, and has demonstrated that an ecosystem of technology companies can work together to measure, manage, and fundamentally change an industry’s approach to energy through innovation. The pace cannot be slackened if we are to deliver the lowest possible impact of our digital lives on the environment.